Appearance
Glossary — one word, one meaning
Every term this skill uses in a fixed sense, and — the expensive half — the pairs that look alike and are not. A word that appears here means exactly this everywhere in the corpus; a word that does not appear is ordinary English. A term enters when a second file needs it; a term used in one file is that file's business.
Confusable pairs — the test tells them apart
| These look alike | …and are not | The test |
|---|---|---|
| update · upgrade | update = new bytes arrive; upgrade = the workspace moves onto them | did anything change here? |
| assign · @-mention | assignment is accountability, and it enqueues a run; a mention of an agent/squad also enqueues a run, of a member or an issue it is free | who is accountable — and does anyone wake? |
| rerun · retry | a CLI issue rerun starts fresh and resets the attempt counter; a per-row retry in the app resumes the session and reuses the working directory | fresh after corrupt state, resume when it's cheaper |
| threshold · cap | a threshold asks before an action above it; a cap stops when a total is reached | before each, or at the sum? |
| grade · model tier | grade is a routing fact in TEAM.md — who gets the work; the tier is what the agent runs on. Grade never enters instructions as an identity | is it about routing, or about compute? |
| project resource (Multica) · shelf (ours) | Multica's is a thing agents work on — a github_repo or a local_directory bound to a project, and a workspace has none; ours is a place you look something up — a STACKS row, _ops/TOOLING.md, sources/SOURCES.md (FLOWS → a shelf serves every flow) | can an agent be dispatched against it, or do you read it? |
Access (Multica) · /multica-ops:mops access | Multica's: who may run this agent (--permission-mode); ours: what a member may direct Mops to do | whose setting is it — the platform's or the methodology's? |
| conductor · Mops | the conductor plans and accepts work (project lead, in the conveyor); Mops advises and builds the machinery (never assigned a task) | does it hold work? |
| crew · quick job | crew = a standing team without a conductor — the owner is the PM; a quick job = 1–2 agents, build → review, none of the machinery, then done | does anything stand after the work ships? |
| consult · research | consult is ephemeral advice, zero standing footprint; /multica-ops:mops research persists cited findings to _ops/research/ | does anything land in the repo? |
| module · skill | a module is an optional capability of the company (design system, brand, theatre); a skill is text an agent loads | toggled for the company, or attached to an agent? |
| workspace · project | the workspace is the company (agents, skills, limits are shared here); a project is a direction inside it (app, site, marketing) | do agents cross it? |
| stage · nesting | --stage N is a number on a sub-issue — a barrier ordering siblings; nesting is issue → sub-issue, exactly two levels | a barrier, or a level? |
| expert · persona | an expert is consulted and cites sources (or is the source, when live); a persona reacts as the audience — direction-only, 🎭-marked, never a hire | does it advise, or simulate? |
| blocked (board) · stuck (run) | blocked is a status someone set, with a reason; stuck is a failed run (agent_error, limits) that rolled the issue back | a decision, or a failure? |
| architecture map · product map | _ops/ARCHITECTURE.md is where the implementation lives — a worker's map of the tree; _ops/MAP.md is how the product is walked — the moves and the things, in the product's own words (/multica-ops:mops map) | does it name files, or moves? |
| verified/recalled/unknown · measured/cited/recalled/judgement | one scale, not two — the three-way labelling is retired. Claims carry measured › cited › recalled › judgement call, plus unknown | — |
| migration · adoption | a migration moves what a workspace already has onto a newer shape; an adoption is taking up something it never used, which a release can make load-bearing. The first is applied on approval; the second is offered with its price and may be declined for good | is there something to convert, or something to start? |
| update · upgrade (bytes vs workspace) | new bytes arriving on the machine is not the workspace moving onto them — swapping the skill files is not migrating the company, and the two are indistinguishable until UPGRADES.md says which happened | did anything in the workspace change? |
Terms with one home
- the four owner-gated kinds — spend · outward · destructive · shape-of-company; route to the owner whoever asks, and no history softens them → SKILL (permissions).
- evidence rung — measured › cited › recalled › judgement call, +
unknown; the rung travels with the claim → SKILL (say what you know). - request — anything that needs a person's answer: an approval, a question, an escalation, a decision. Carries an age and, where knowable, what the wait costs; a report is where findings go to die → SKILL · PLAYBOOKS (status shape).
- grant —
right · grantee · scope · duration: the only way a gate loosens; visible while it lives, expiry evaluated at the gate check → PLAYBOOKS (Gates). enforced_by—request·validator·git-host·platform·prose-only; every gate declares its value honestly → PLAYBOOKS (Gates).- waste slices — outcome · attempt · tier · theatre/system: what was spent on work that produced nothing, derived from the same ledger as the spend → PLAYBOOKS (ledger).
- fingerprint — the per-class hash of workspace state in
_ops/.workspace-state.json; what answers "what changed that we didn't change" → PLAYBOOKS (drift). - conveyor — the pull-based flow: conductor seeds, leaders route, stage barriers sequence, @-mention hands off → SKILL.
- theatre — the personas module at full size: staged personas, bias profiles, mixed live+synthetic rounds, direction-only verdicts → MODULES.